IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 62, NO.
7, SEPTEMBER 2013
Optimizing Electric Vehicle Charging:
A Customers Perspective
Chenrui Jin, Student Member, IEEE, Jian Tang, Member, IEEE, and Prasanta Ghosh, Senior Member, IEEE
AbstractElectric vehicles (EVs) are considered to be a promis-
ing solution for current gas shortage and emission problems.
To maximize the benefits of using EVs, regulated and optimized
charging control needs to be provided by load aggregators for
connected vehicles. An EV charging network is a typical cyber-
physical system, which includes a power grid and a large num-
ber of EVs and aggregators that collect information and control
the charging procedure. In this paper, we studied EV charging
scheduling problems from a customers perspective by jointly con-
sidering the aggregators revenue and customers demands and
costs. We considered two charging scenarios: static and dynamic.
In the static charging scenario, customers charging demands are
provided to the aggregator in advance; however, in the dynamic
charging scenario, an EV may come and leave at any time, which
is not known to the aggregator in advance. We present linear
programming (LP)-based optimal schemes for the static problems
and effective heuristic algorithms for the dynamic problems. The
dynamic scenario is more realistic; however, the solutions to the
static problems can be used to show potential revenue gains and
cost savings that can be brought by regulated charging and, thus,
can serve as a benchmark for performance evaluation. It has been
shown by extensive simulation results based on real electricity
price and load data that significant revenue gains and cost savings
can be achieved by optimal charging scheduling compared with
an unregulated baseline approach, and moreover, the proposed
dynamic charging scheduling schemes provide close-to-optimal
solutions.
Index TermsCharging regulation, electric vehicle (EV), opti-
mization, smart grid.
I. I NTRODUCTION
A S THE shortage of petroleum storage and the increase
in CO2 , SO2 , and NOx emissions receive increasing
attention, policy makers, engineers, and business leaders are
searching for alternative energy sources, which are both eco-
nomically and environmentally friendly [23]. Using electric
vehicles (EVs) instead of traditional internal combustion engine
vehicles is considered as a promising solution. Compared with
traditional vehicles, EVs can offer many benefits such as lower
operational costs, lower gas emissions, and so on [6]. Renew-
able energy, such as solar and wind power, can be also utilized
Manuscript received February 6, 2012; revised July 13, 2012 and October 16,
2012; accepted January 29, 2013. Date of publication March 7, 2013; date of
current version September 11, 2013. This work was supported in part by the
Department of Energy under Grant DE-OE0000495 and in part by the National
Science Foundation under Grant CNS-1113398. The review of this paper was
coordinated by Prof. J. Wang.
The authors are with the Department of Electrical Engineering and Computer
Science, Syracuse University, Syracuse, NY 13244 USA (e-mail: cjin01@
syr.edu; jtang02@syr.edu; pkghosh@syr.edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2013.2251023
0018-9545 2013 IEEE
2919
to charge EVs to further reduce costs. However, integrating EVs
into the power grid is very challenging. Unregulated charging of
EVs with fast charging speeds can result in a heavy load burden
on the power grid system and may even cause the system to
break down [33]. The high cost of initial investment on EV
purchase is also a concern for consumers, although this can
be compensated by government incentive funds and relatively
lower power costs. To maximize the benefits of using EVs, we
need regulated and optimized charging control strategies. Indi-
vidual customers are usually not able to perform the charging
regulation to their own cars in the most efficient way, although
they have the motivation to save on charging costs. They can
provide their needs for charging to load aggregators, which act
as a control interface between consumers and the grid operator
to provide regulated charging for connected vehicles with joint
consideration for benefits of both consumers and the grid.
An EV charging network is a typical cyber-physical system,
which includes a power grid and a large number of EVs and
aggregators that collect information and control the charging
procedure. The development of modern sensing and commu-
nication technologies enables the energy management system
to obtain information from the power grid system and energy
consumers efficiently. New algorithms and automatic operation
strategies are needed for more precise and efficient control
to enable intelligent load aggregation, reduce electricity costs,
prevent the system from overloading, and satisfy customers
demands. An aggregator can perform centralized integration
and control for EV charging. Usually, EVs can be charged at
various charging rates. The aggregator automatically controls
both the charging rate and the regulation capacity for the
connected EVs through power electronics.
The revenue of aggregators consists of two parts: the mark
over price for energy sold to customers and the regulation ser-
vice provided to the power grid. EV charging optimization has
been studied by a few recent works [18], [34], most of which,
however, focused on maximizing the revenue of an aggregator
without carefully addressing customers needs. Specifically, the
existing methods [18], [34] may not necessarily lead to the
maximum benefit for consumers, i.e., the minimum charging
cost. Moreover, at the end of charging, some EVs may not be
fully charged or charged to desired states of charge (SOCs).
In this paper, we consider EV charging from a customers
perspective and present optimal schemes to maximize the rev-
enue of aggregators and minimize the total charging cost of
customers, which allow customers to specify their charging
demands (starting time, finishing time, desired SOC, etc.).
Meanwhile, we also take into account the requirements of the
grid system (such as the power delivery capacity requirement)
2920 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 62, NO. 7, SEPTEMBER 2013
to ensure that it operates normally. We consider two charging
scenarios: static and dynamic. In the static charging scenario,
customers charging demands are provided to the aggregator
in advance; however, in the dynamic charging scenario, an
EV may come and leave at any time, which is not known
to the aggregator in advance. The dynamic scenario is more
realistic; however, the solutions to the static problems can be
used to show potential revenue gains and cost savings that can
be brought by regulated charging. Moreover, they can serve as
a benchmark for performance evaluation. To the best of our
knowledge, we are the first to conduct a comprehensive study
for EV charging from a customers perspective with emphases
on different individual demands and a tradeoff between the
aggregators revenue and charging costs. We summarize our
contributions as follows.
1) We present optimal schemes for the static charging
scheduling problems and heuristic algorithms for the
dynamic problems with considerations for both the ag-
gregators revenue and customers demands and cost.
2) We present extensive simulation results based on real
electricity price and load data to show the potential
revenue gains and cost savings that can be brought by
optimal charging scheduling and the performance of the
proposed dynamic charging scheduling algorithms.
II. R ELATED W ORK
EV charging has attracted substantial research attention
due to its potential impact on the grid system. Experimental
results from the Oak Ridge National Laboratory [17] and the
National Renewable Energy Laboratory [13] showed that in
most regions, additional generation capacities are needed to
meet demands of EVs if they are charged with an uncontrolled
charging strategy. Researchers from Virginia Tech, Blacksburg,
VA, USA, pointed out in [31] that as EVs take a greater share in
the fleet market, they may bring potential challenges to electric
utilities, particularly at the distribution level. Their simulation
results showed that at the EV penetration level of interest,
new load peaks will be created, which, in some cases, may
exceed the distribution transformers capacities. EV stagger
charge and household load control were exploited to solve this
problem [31]. In [15], Green et al. concluded that the impact
of EV charging on distribution networks can be determined by
the following factors: driving patterns, charging characteristics,
charging timing, and vehicle penetration. A dynamic program-
ming model for assessing the impact of EVs on the distribution
grid of Belgium was developed by Clement et al. [9].
The optimal operation of an aggregator for controlled EV
charging has been studied by a few recent works. In [18],
Han et al. designed an aggregator that makes efficient use
of the distributed power of EVs to produce the desired grid-
scale power. Both the cost arising from battery charging and
the revenue obtained by providing the regulation service were
considered. A dynamic programming algorithm was presented
to compute the optimal charging control for each vehicle.
Sortomme and El-Sharkawi [34] presented algorithms to find
the optimal charging rates with the objective of maximizing
the aggregators profit. Pedrasa et al. [28] improved the ba-
sic formulation of cooperative particle swarm optimization by
introducing stochastic repulsion among particles to investigate
the potential consumer value added by coordinated distributed
energy resource scheduling. Jang et al. [21] proposed a method
for analytic estimation of the probability distribution of the pro-
cured power capacity to obtain an optimal contract size regard-
ing the frequency of regulation. Deilami et al. [12] proposed a
real-time smart load management control strategy to minimize
the total cost of energy generation and associated grid energy
losses, which utilizes the maximum sensitivities selection op-
timization technique. In a recent work [11], Dallinger et al.
presented a new approach to analyze the economic impacts of
vehicle-to-grid (V2G) regulation and performed a case study
for Germany using average daily data.
In [27], Ma et al. developed a decentralized method to
coordinate the charging of autonomous plug-in EVs using
concepts from noncooperative games, for the case where central
computing resources or communications infrastructure are not
available or adequate. In [7] and [20], conceptual frameworks
for actively involving highly distributed loads in system control
actions are presented, and the communications infrastructure
required to support such a load control scheme is discussed.
In [8], Callaway developed new methods to model and control
the aggregated power demand from a population of thermostat-
ically controlled loads, with the goal of delivering services such
as regulation and load following. In [22], Kempton and Tomic
examined the systems and processes needed to tap energy in ve-
hicles and implement V2G. In [29], an optimization algorithm
was presented to manage a virtual power plant composed of
a large number of customers with thermostatically controlled
appliances.
The differences between our work and these related works
are summarized as follows: 1) Unlike papers addressing only
the aggregators benefit [11], [18], [21], [34], we consider the
charging problems from a customers perspective and aim at
satisfying customers demands and reducing their costs. 2) We
conduct a comprehensive study for the EV charging problems
by considering both the static and dynamic cases. The methods
proposed for the dynamic problems can be used in a real-
time manner. However, algorithms in [9], [13], [15], [31], and
[35] are not real-time algorithms. 3) Our schemes determine
the charging rate for each individual EV, whereas in [31] and
[35], the EV charging pool was treated as a whole operation
unit, and the charging rate is assigned to the unit based on
its statistical behavior. 4) Closely related works [11][13],
[17], [28] presented heuristic algorithms that cannot provide
any performance guarantees. However, we present schemes to
produce optimal solutions for the static charging problems.
5) Regulation services provided by EV charging control have
not been considered in [27]. In [20] and [22], the authors
presented conceptual frameworks without presenting specific
optimization algorithms, whereas in our work, we presented
algorithms to optimize EV charging scheduling with consid-
eration for customer demands. Both [8] and [29] studied ther-
mostatically controlled loads instead of EV charging loads;
thus, the problems in their works are mathematically different
from ours.
JIN et al.: OPTIMIZING ELECTRIC VEHICLE CHARGING: A CUSTOMERS PERSPECTIVE
Fig. 1. EV charging network.
III. S YSTEM M ODEL
Here, we introduce the system model and necessary
notations.
An EV charging network consists of a substation, transform-
ers connected by transmission lines, EVs connected to the leaf-
node transformers, and an aggregator, which is shown in Fig. 1.
In such a network, the aggregator serves as a central control
node that collects information from both the power grid and
connected EVs and instructs the grid to charge each EV with
a charging rate given by the charging scheduling algorithm in
each hour. The efficient and safe delivery of power from the
substation to EVs requires that, at any time, the power flow
through each transformer or each branch does not exceed its
delivery capacity. Usually, if the power flow through each leaf-
node transformer does not exceed its delivery capacity, all the
other devices and transmission lines will not be overloaded.
This implies that the power delivery through one leaf-node
transformer is independent from that through another leaf-
node transformer. Therefore, we can divide EVs into multiple
groups such that each group includes EVs associated with a
leaf-node transformer and solve a charging scheduling problem
independently for each group. This way, we can significantly
reduce the computational complexity of the problem. In the
following, we will focus on solving the charging scheduling
problems for EVs in one group. In addition, in each transformer,
in addition to power consumed by EVs, there are basic daily
power loads contributed by all other electronic appliances (e.g.,
refrigerators, coffee makers, washing machines, etc.), which
can be estimated from historical data and should be counted
toward the total power load. This is referred to as base load in
the following.
An EV can be connected to or disconnected from the charg-
ing network at any time according to the customers needs. A
customer will input his desired finishing time and final SOC of
the battery for his car when it is connected to the network. Each
charging task can be characterized by a 5-tuple (i, si , fi , ei , ei ),
where i is the vehicle ID, si is the starting time, fi is the
desired finishing time, ei is the initial SOC of the battery,
and ei is the desired SOC after charging. Since most charging
tasks will be undertaken at night, we consider a charging
scheduling period starting from 12:00 P. M . (noon) and ending at
12:00 P. M . (noon) on the next day. Moreover, for simplicity,
we use a single decimal number between 0 and 24 to present
2921
Fig. 2. Daily electricity wholesale prices in Central New York.
a time, i.e., hour:minute. For example, suppose that an EV i is
connected to the grid at 6:15 P. M . with an initial SOC of 0.6.
It is scheduled to leave at 7:30 A . M . on the next day, with the
battery fully charged. Then, the corresponding charging task
can be presented as (i, 6.25, 19.5, 0.6, 1).
An aggregator is able to charge each EV at any charging rate,
ranging from a minimum value to a maximum value, depending
on the type and the condition of the EVs battery [22], [36].
When EVs are connected to the grid, the aggregator controls
charging rates and makes sure that the total charging rate of EVs
associated with each transformer does not exceed its delivery
capacity. This is referred to as the power delivery capacity
constraint.
In addition, the aggregator will also try to maximize its
revenue. The aggregator buys electricity from the grid at a
wholesale price, which usually varies on an hour-to-hour basis,
as shown in Fig. 2. We plotted this figure based on real prices
in Central New York during July 1420, 2011, obtained from
the New York Independent System Operator (NYISO), which
is a regional transmission organization in North America [25].
In this figure, we can see that the daily wholesale electricity
price follows a quite similar hourly change pattern on different
days, i.e., higher values at peak hours (such as 3 P. M .6 P. M .)
and lower values at off-peak hours (such as 3 A . M .5 A . M .).
At a particular hour, the day-to-day price difference is not
significant compared with the hourly difference. The aggre-
gator will add a markup price to the wholesale price and sell
electricity to customers for charging their vehicles. Note that
we consider a widely used pricing model in which the power
grid operator specifies electricity prices on an hourly basis,
which do not change with vehicle charging and discharging
policies.
The power grid operator prefers relatively even load distribu-
tion and stable frequency for the purpose of system stability
and reliability. The aggregator can adjust charging rates of
connected EVs to produce relatively even load distribution, i.e.,
charge EVs with higher rates when the base load is low, and
vice versa. Moreover, when the grid calls for extra load or
2922 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 62, NO. 7, SEPTEMBER 2013
Fig. 3. Regulation capacities provided by a single EV.
additional power supply in case of emergency, the aggregator
can respond with temporally increased or temporally decreased
charging rates. This is referred to as regulation service in
the electricity market. The amount of charging rate that an
aggregator can increase or decrease as requested by the grid is
referred to as regulation capacity. The aggregator determines
the regulation capacity offered to the grid while performing
charging services. The calculation of regulation capacity that
one EV can provide to the grid is shown in Fig. 3. In the
figure, the potential regulation capacity that can be provided
by each EV is limited by the operational range of charging
rates. The difference between the upper charging rate limit
or the maximum charging rate limit (whichever is smaller) and
the scheduled charging rate is the regulation-down capacity.
The upper charging rate limit is given by the battery of the
EV, which is the largest rate at which the battery can draw
power. The maximum charging rate limit is the power needed
to charge the EV to its desired SOC in one timeslot, which can
be calculated using the following equation:
(ei xi,t ) Ci
Pi,t = i, t (1)
Ei
where ei is the desired SOC of EV i, xi,t is the current
SOC of EV i at the beginning of timeslot t, Ci is the battery
capacity, and Ei is the battery charging efficiency of EV
i (the ratio between the effectively stored energy and input
energy). Obviously, the charging rate of each EV cannot exceed
this value in every timeslot. In addition, when charging a
group of EVs, the total regulation-down capacity should be
further restricted by the delivery capacity of the transformer.
Specifically, the aggregator determines charging rate pi,t for
each connected EV to make sure that in each timeslot t, the
total charging rate plus the regulation-down
 capacity is within
its upper operational limit, i.e., i pi,t + i di,t R, where
di,t is the regulation-down capacity, and R is the transformer
delivery capacity. This way, the aggregator can help keep the
system frequency stable, which is why the power grid operator
is willing to pay for the regulation service. The regulation-
up capacity, on the contrary, is the ability that the EV can
decrease its charging rate to help boost the system frequency
and is given by the difference between the lower charging rate
limit and the scheduled charging rate of the EV (as shown in
Fig. 3). The regulation capacity is the sum of the regulation-up
capacity and the regulation-down capacity. The aggregator will
be paid by the power grid operator based on the regulation
capacities. We use a model similar to that in [34] to calculate
the aggregators revenue. On one hand, an aggregator makes
money from customers via the price difference between retail
and wholesale. On the other hand, its revenue comes from the
grid operator for its regulation service. Equation (2) is used to
calculate the aggregators revenue obtained from an EV in a
timeslot, i.e.,
mi,t = at ri,t + M pi,t hi,t (2)
where mi,t is the aggregators revenue from EV i in timeslot
t, at is the regulation price in timeslot t, ri,t is the regulation
capacity from EV i in timeslot t, M is the markup price, pi,t
is the charging rate for EV i in timeslot t, and hi,t specifies for
how long EV i is connected to the grid during timeslot t. Note
that the aggregator makes money from the grid operator only
for EV charging but not for base loads.
From a customers perspective, the cost of charging an EV in
a timeslot is given as follows:
ci,t = (M + gt )pi,t hi,t (3)
where ci,t is the charging cost for charging EV i in timeslot t,
and gt is the wholesale electricity price in timeslot t. Note that
if an EV is connected to the charging network in the middle of
a timeslot, it does not pay for the period during which it is not
connected during that timeslot.
IV. P ROBLEM D EFINITION
The electricity market operates on an hourly basis; there-
fore, we can naturally divide the time domain into timeslots
with a duration of 1 h each and consider T = 24 h as a
charging period, starting at 12:00 P. M . (noon). We define a
charging schedule for a given charging task i as a vector i =
[pi,1 , . . . , pi,t , . . . , pi,T ] in which each entry specifies the charg-
ing rate at hour t. We say that a charging schedule is feasible
if both customers demands (specified by charging tasks) and
the system loading requirement are satisfied. In addition, we
consider two objectives: maximizing the aggregators revenue
and minimizing the total electricity cost of customers. More-
over, we take account of customers costs when maximizing
the aggregators revenue, i.e., for the revenue maximization
problems; we also make sure that the cost of a charging task
does not exceed a given upper bound. This way, we can achieve
a tradeoff between the aggregators revenue and the customers
costs. Accordingly, a charging schedule can be defined for a
set of given tasks: = {i : i {1, . . . , N }}, where N is the
number of charging tasks.
We study two charging scenarios: static and dynamic. In the
static version, we assume that the aggregator knows all the
scheduling tasks (starting time, finishing time, initial SOC, and
JIN et al.: OPTIMIZING ELECTRIC VEHICLE CHARGING: A CUSTOMERS PERSPECTIVE 2923

expected SOC) in a coming scheduling period beforehand. The TABLE I

M AJOR N OTATIONS
aggregator calculates a charging schedule using an algorithm,
and each vehicle will follow it for charging after its arrival.
In reality, information on some charging tasks may not be
available ahead of time. Therefore, we consider a more realistic
dynamic scenario in which the aggregator is not assumed to
know the charging tasks before EVs actually arrive. A charging
scheduling algorithm will be used by the aggregator to calculate
the charging schedule for each vehicle right after it is connected
to the grid.
First, we define the static scheduling problems. Given a set
of charging tasks Q = {(1, s1 , f1 , e1 ), . . . , (i, si , fi , ei ), . . . ,
(N, sN , fN , eN )}; base loads l = [l1 , . . . , lt , . . . , lT ]; elec-
tricity prices g = [g1 , . . . , gt , . . . , gT ]; regulation prices a =
[a1 , . . . , at , . . . , aT ]; transformer delivery capacity R; EV bat-
tery capacities C = [C1 , . . . , Ci , . . . , CN ]; EV battery charging
efficiency E = [E1 , . . . , Ei , . . . , EN ], which is the ratio be-
tween the effectively stored energy and input energy; and upper
and lower charging rate limits P = [P1 , . . . , Pi , . . . , PT ] and
P = [P1 , . . . , Pi , . . . , PT ], we define the following optimiza-
tion problems.
Definition 1: The Maximum Revenue Static Charging
Scheduling Problem (R-SCSP) seeks a feasible charging
schedule for the set of given chargingtasks Q such that
T N
the corresponding aggregators revenue t=1 i=1 mi,t is
maximum among all feasible charging schedules.
In addition, we also aimed at minimizing the total charging
cost from a customers perspective.
Definition 2: The Minimum Cost Static Charging
Scheduling Problem (C-SCSP) seeks a feasible charging
schedule for the set of given chargingtasks  Q such that
t=T i=N
the corresponding total electricity cost t=1 t=1 ci,t is
minimum among all feasible charging schedules.
The dynamic charging scheduling problems are the same as
their static counterparts except that the charging task sequence
is not assumed to be known in advance. Correspondingly, we
call the two problems as the Maximum Revenue Dynamic subject to
Charging Scheduling Problem (R-DCSP) and the Minimum
Cost Dynamic Charging Scheduling Problem (C-DCSP).
ei i, t = si 
For easy reference, we summarize all major notations in 
xi,t = ei i, t = fi (5)
Table I. Ei hi,t1 pi,t1
xi,t1 + Ci , otherwise

Pi pi,t Pi i, t (6)

V. P ROPOSED E LECTRIC V EHICLE C HARGING S CHEMES
 
Here, we present optimal schemes for the problems defined 
T 
N

in the last section. We find that the static problems can be (M + gt ) pi,t hi,t B (7)
t=1 i=1
formulated as linear programming (LP) problems, which are
known to be solvable in polynomial time [4]. For the dynamic ri,t = ui,t + di,t i, t (8)
problems, we present polynomial-time heuristic algorithms.

pi,t Pi i, t s.t. hi,t = 1
ui,t = (9)
0, otherwise
A. Static Charging Scheduling

First, we present the LP formulation for the R-SCSP.LP- min{Pi,t , Pi } pi,t i, t s.t. hi,t = 1
di,t = (10)
R-SCSP: 0, otherwise
N 
T  T  N 
N 
N
max at ri,t + M pi,t hi,t (4) pi,t + lt + di,t R t. (11)
xi,t ,pi,t ,ri,t
t=1 i=1 t=1 i=1 i i
2924 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 62, NO. 7, SEPTEMBER 2013

In this formulation, (5) assigns a value for the SOC of that previously presented since it is used for a single EV i.
each connected EV in each hour properly. This way, each EV ThusLP-R-DCSP (i)
is guaranteed to be charged to the desired SOC when it is
disconnected from the charging network. Note that in these 
T 
T
max at ri,t + M pi,t hi,t (14)
equations, hi,t gives the actual connection time of each EV xi,t ,pi,t ,ri,t
t=1 t=1
during each timeslot. An EV may not always arrive at the
beginning of a timeslot and leave at the end of a timeslot. subject to
Therefore, in a timeslot, the connection time of an EV may be
less than 1 h, and this would affect the calculation of regulation ei t = si 

capacities, charging energy, and cost. Note that hi,t is NOT a x t = ei t = fi (15)
Ei hi,t1 pi,t1
decision variable, and its value can be precalculated using the xi,t1 + Ci , otherwise
following equation once a charging task is given: Pi pi,t Pi t (16)
T

1, si  < t < fi  (M + gt )pi,t hi,t B (17)

1, t = si , si = si  t=1
hi,t = si si , t = si , si
= si  (12) ri,t = ui,t + di,t t (18)


fi fi ,
t = fi , fi
= fi  
pi,t Pi t s.t. hi,t = 1
0, otherwise. ui,t = (19)
0, otherwise

Constraint (6) ensures that in each timeslot, the charging min{Pi,t , Pi } pt t s.t. hi,t = 1
di,t = (20)
rate of each EV is no smaller than its lower limit but no 0, otherwise
larger than its upper limit. In addition, the total charging cost  i  i

of an EV should not exceed cost upper bound B, which is pj,t + lt + di,t R t. (21)
j=1 j=1
guaranteed by constraint (7). B is the maximum total cost that
customers are willing to pay for charging their EVs, which is In this formulation, the objective function and constraints are
specified by customers based on their needs. We make these similar to those in the R-SCSP. However, we only calculate
constraints general enough such that the cost upper bound can the charging scheduling for the newly arriving EV i. Note that
be set to any reasonable value based on various factors such constraints (17) and (21) are different from (7) and (11) because
as emergency level, the cost without regulation, the minimum we only consider this new EV i and all those EVs that were
possible total cost, etc. Constraints (8)(11) use the model de- connected before EV i. Similarly, for the C-DCSP, we solve the
scribed in Section IV to calculate the regulation capacities and following LP:LP-C-DCSP(i):
make sure that there is no violation on the transformer delivery
capacity. Note that since the regulation service is provided on 
T

an hourly basis, the regulation capacity should be consistent min (M + gt )pi,t hi,t (22)
xi,t ,pi,t ,ri,t
during the whole timeslot. In the hours when the EV arrives or t=1

leaves in the middle, regulation service is not considered to be

provided by the aggregator. Hence, (9) and (10) ensure that the
regulation capacity will NOT be accounted for in those hours.
ThusLP-C-SCSP:
 

T 
N
min (M + gt ) pi,t hi,t
xi,t ,pi,t ,ri,t
t=1 i=1
subject to constraints (5)(6) and (8)(11).
B. Dynamic Charging Scheduling
Either the R-DCSP or the C-DCSP cannot be solved by
solving a single LP problem because the charging tasks are
not known in advance. We propose two schemes to solve the
dynamic scheduling problems. In the first scheme, we solve an
LP problem to find charging schedule i every time charging
task i arrives (i.e., an EV is connected to the charging net-
work). Once the charging schedule for an EV is calculated, it
will not be changed during the whole charging period. The
LP problem is presented in the following, which has a much
smaller size (the number of constraints and variables) than
subject to constraints (15)(21).
Although this approach can find a charging schedule with
either the maximum incremental revenue (LP-R-DCSP) or the
minimum incremental cost (LP-C-DCSP) every time an EV is
connected to the network, it cannot guarantee that the solution
is optimal for either the R-DCSP or C-DCSP. To improve its
(13)
performance further, we come up with another scheme, namely,
dynamic charging scheduling with updating, which adjusts the
charging schedules for the connected EVs every time k more
EVs arrive. This can be done by solving an LP problem that
is the same as the LP-R-SCSP or LP-C-SCSP, except that
the input only includes EVs that are currently connected to
the network (instead of all EVs). The smaller the value of k,
usually, the better the performance, but the higher the overhead.
VI. S IMULATION R ESULTS
We evaluated the performance of the proposed schemes
using real electricity prices and real base load and battery
data. Specifically, the hourly electricity prices and regulation
prices in each day in Central New York were obtained from the
NYISO [25]. Average electricity prices and regulation prices
over 30 days (July 130, 2011) were used for simulation.
JIN et al.: OPTIMIZING ELECTRIC VEHICLE CHARGING: A CUSTOMERS PERSPECTIVE 2925

TABLE II
C OMMON S IMULATION S ETTINGS

The aggregator charges a markup price (M ) of $0.05 per

kilowatthour from customers. We set EV-battery-related param-
eters, including the charging rate limit and battery capacity Fig. 4. Revenue gains given by optimal charging scheduling.
based on the specifications of the Chevy Volt Li-ion battery
[3]. A typical summer base load profile for a 100-household
residential community was generated using the demand profile
generators from the University of Strathclyde, Glasgow,
U.K. [26].
In the simulation, N charging tasks were generated for a
scheduling period from 12 P. M . (noon) to 12 P. M . on the next
day to simulate the overnight EV charging. According to the
survey data in [14], we used Gaussian distributions to model
vehicles travel pattern and generate EV arrival (starting) and
departure (finishing) times. Specifically, the starting times si
follow a normal distribution with a mean of = 6 P. M . and a
standard deviation of = 2 h; the desired finishing times fi
follow a normal distribution with = 7 A . M . and = 2 h; and
the daily travel distances follow a lognormal distribution with Fig. 5. Cost savings given by optimal charging scheduling.
= 3.22 mi and = 0.66 mi. The initial SOCs ei were set to
total charging cost (given by the C-SCSP scheme) with the
be distributed in the range [0.3, 0.9], and the desired SOC ei
baseline solutions. In the simulation, we fixed the EV battery
was set to 0.9 for each EV. A total of ten seeds (from one
upper charging rate limit Pi at 4.4 kW and the charging cost
to ten) were used to generate ten sets of random input data,
upper bound at the total cost of unregulated charging. More-
and the average values over ten runs were presented as results
over, we perform simulation runs on networks with sizes of
in the following figures. The common simulation settings are
N = {10, 20, 50, 100, 200}, respectively.
summarized in Table II.
In Figs. 4 and 5, we can see that significant revenue gains
In the simulation, an unregulated charging scheme was used
and cost savings can be brought by optimal charging schedul-
as a baseline solution for comparison, in which the aggregator
ing. Specifically, on average, the aggregators revenue can be
serves EVs on a first-come first-serve basis and charge EVs with
improved by 132%, and the total charging cost can be reduced
the maximal allowable charging rate. We used the aggregators
by 17.4%, compared with unregulated charging.
revenue, the total charging cost as the performance metrics. The
following parameters may have a significant impact on system
performance: the number of EVs (N ), the total charging cost B. Performance of Dynamic Charging Scheduling Schemes
upper bound (B), and the upper charging rate limit (Pi ). We
As previously mentioned, the dynamic charging scheduling
first try to find out how they can affect the system performance
schemes are more likely to be used in reality. To evaluate their
using the optimal static charging scheduling schemes by setting
performance, we used the maximum possible aggregator rev-
them to different values. As previously mentioned, the optimal
enues given by the R-SCSP scheme, the minimum possible total
solutions given by the static charging scheduling schemes can
charging costs given by the C-SCSP scheme, and the revenues
be used to show the potential benefit that can be brought by
and costs given by the baseline approach as benchmarks. The
regulated charging and serve as a benchmark for performance
simulation settings in the scenario were the same as those in
evaluation. We used the CPLEX [10] to solve all the LP
the first scenario previously described, except that the charging
problems.
cost upper bound was set as the minimum possible total cost.
The simulation results are presented in Figs. 6 and 7.
A. Potential Benefits Given by Optimal Charging Scheduling
In Figs. 6 and 7, we can see that the dynamic charg-
To show the potential benefits that can be brought by optimal ing scheduling schemes with updating (R-DCSP-updating and
charging scheduling, we compare the maximum aggregators C-DCSP-updating) consistently perform better than the dy-
revenue (given by the R-SCSP scheme) and the minimum namic scheduling schemes without updating (R-DCSP and
2926
Fig. 6. Performance of dynamic charging scheduling schemes in terms of
revenue.
Fig. 7. Performance of dynamic charging scheduling schemes in terms of
cost.
C-DCSP). Hence, the updating procedure can actually improve
system performance. Moreover, on average, the aggregators
revenue given by the R-DCSP-updating scheme is only 7%
away from the optimal values, and the total charging cost
given by the C-DCSP-updating scheme is almost the same as
the optimal values. Therefore, similar to the optimal charging
scheduling scheme, the R-DCSP-updating scheme improves
the aggregators revenue by 113% and reduces the total charg-
ing cost by 17.4% on average, compared with the baseline
approach. Therefore, we conclude that the R-DCSP-updating
scheme provides close-to-optimal solutions and achieves a
good tradeoff between revenue and cost.
C. Impact of the Upper Charging Rate Limit on
System Performance
The impact of the upper charging rate limit on the maximum
aggregators revenue and the minimum total charging cost is
shown in Figs. 8 and 9. We performed simulation runs on a
network with N = 200 EVs by changing the upper charging
rate limit from 3.3 to 8.8 kW with a step size of 1.1 kW. The cost
upper bound was set to the total cost of unregulated charging.
The maximum aggregators revenues were computed by the
R-SCSP scheme, and the minimum total charging costs were
calculated by the C-SCSP scheme.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 62, NO. 7, SEPTEMBER 2013
Fig. 8. Maximum aggregators revenue versus the upper charging rate limit.
Fig. 9. Minimum total charging cost versus the upper charging rate limit.
We make the following observations from the simulation
results: The upgrade of the upper charging rate limit does not
benefit the customers much (i.e., does not significantly reduce
the charging costs). However, the upper charging rate limit
has a positive impact on the aggregators revenue, i.e., the
aggregators revenue increases with the upper charging rate
limit. This is because the upgrade of this limit improves the
regulation capacity, which determines the revenue. However,
this increase can only be achieved by upgrading the charging
station hardware. In addition, when the upper limit becomes
relatively large, the increase in revenue slows down, because
there exists a fixed transformer delivery capacity R.
VII. C ONCLUSION
In this paper, we have studied EV charging scheduling prob-
lems from a customers perspective by jointly considering the
aggregators revenue and customers demands and costs. We
presented LP-based optimal schemes for the static charging
scheduling problems and effective heuristic algorithms for the
dynamic problems. It has been shown by extensive simulation
results based on real electricity price and load data that, com-
pared with an unregulated baseline approach, the aggregators
revenue can be improved by 113%, and the total charging cost
can be reduced by 17.4% on average. Moreover, the proposed
dynamic charging scheduling schemes provide solutions that
are very close to optimal.
JIN et al.: OPTIMIZING ELECTRIC VEHICLE CHARGING: A CUSTOMERS PERSPECTIVE
R EFERENCES
[1] S. L. Andersson, A. K. Elofsson, M. D. Galus, L. Gransson, S. Karlsson,
F. Johnsson, and G. Andersson, Plug-in hybrid electric vehicles as reg-
ulating power providers: Case studies of Sweden and Germany, Energy
Policy, vol. 38, no. 6, pp. 27512762, Jun. 2010.
[2] S. Bashash and H. K. Fathy, Transport-based load modelling and sliding
mode control of plug-in electric vehicles for robust renewable power
tracking, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 526534, Mar. 2012.
[3] Chevy Volt Specifications. [Online]. Available: http://www.chevrolet.
com/volt
[4] M. S. Bazaraa, J. J. Jarvis, and H. D. Sherali, Linear Programming and
Network Flows, 3rd ed. Hoboken, NJ, USA: Wiley, 2005.
[5] A. Brooks and S. H. Thesen, PG & E and Tesla motors: Vehicle to grid
demonstration and evaluation program, in Proc. Int. Battery, Hybrid Fuel
Cell Elect. Veh. Symp. Exhib., 2007, pp. 110.
[6] Air Resources Board, Research Division, Regulation, California Environ.
Protect. Agency, Sacramento, CA, USA, 2010.
[7] D. Callaway and I. Hiskens, Achieving controllability of electric loads,
Proc. IEEE, vol. 99, no. 1, pp. 184199, Jan. 2011.
[8] D. S. Callaway, Tapping the energy storage potential in electric loads to
deliver load following and regulation, with application to wind energy,
Energy Convers. Manage., vol. 50, no. 5, pp. 13891400, May 2009.
[9] K. Clement, E. Haesen, and J. Driesen, The impact of charging plug-in
hybrid electric vehicles on a residential distribution grid, IEEE Trans.
Power Syst., vol. 25, no. 1, pp. 371380, Feb. 2010.
[10] ILOG Software Inc., CPLEX 10.1. [Online]. Available: http://www.ilog.
com
[11] D. Dallinger, D. Krampe, and M. Wietschel, Vehicle-to-grid regulation
reserves based on a dynamic simulation of mobility behavior, IEEE
Trans. Smart Grid, vol. 2, no. 2, pp. 302313, Jun. 2011.
[12] S. Deilami, A. S. Masoum, P. S. Moses, and A. S. Mohammad, Real-
time coordination of plug-in electric vehicle charging in smart grids to
minimize power losses and improve voltage profile, IEEE Trans. Smart
Grid, vol. 2, no. 3, pp. 456467, Sep. 2011.
[13] P. Denholm and W. Short, An evaluation of utility system impacts and
benefits of optimally dispatched plug-in hybrid electric vehicles, Nat.
Renew. Energy Lab., Golden, CO, USA, Tech. Rep., 2006.
[14] Summary of Travel Trends, 2009 National Household Travel Survey, U.S.
Dept. Transp., Washington, DC, USA, 2009.
[15] R. C. Green, L. Wang, and M. Alam, The impact of plug-in hybrid
electric vehicles on distribution networks: A review and outlook, Renew.
Sustain. Energy Rev., vol. 15, no. 1, pp. 544553, Jan. 2011.
[16] C. Guille and G. Gross, A conceptual framework for the vehicle-to-grid
(V2G) implementation, Energy Policy, vol. 37, no. 11, pp. 43794390,
Nov. 2009.
[17] S. W. Hadley, Impact of plug-in hybrid vehicles on the electric grid, Oak
Ridge Nat. Lab., Oak Ridge, TN, USA, Tech. Rep. [Online]. Available:
http://www.ornl.info/info/ornlreview/v40_2_07/2007_plug-in_paper.pdf
[18] S. Han, S. Han, and K. Sezaki, Development of an optimal vehicle-to-
grid aggregator for frequency regulation, IEEE Trans. Smart Grid, vol. 1,
no. 1, pp. 6572, Jun. 2010.
[19] A. Hajimiragha, C. A. Caizares, M. W. Fowler, and A. Elkamel, Optimal
transition to plug-in hybrid electric vehicles in Ontario, Canada, consid-
ering the electricity-grid limitations, IEEE Trans. Ind. Electron., vol. 57,
no. 2, pp. 690701, Feb. 2010.
[20] I. Hiskens and D. Callaway, Achieving controllability of plug-in electric
vehicles, in Proc. Veh. Power Propul. Conf., 2009, pp. 12151220.
[21] S. Jang, S. Han, S. H. Han, and K. Sezaki, Optimal decision on contract
size for V2G aggregator regarding frequency regulation, in Proc. Int.
Conf. Optim. Elect. Electron. Equip., 2010, pp. 5462.
[22] W. Kempton and J. Tomic, Vehicle-to-grid power implementation: From
stabilizing the grid to supporting large-scale renewable energy, J. Power
Sources, vol. 144, no. 1, pp. 280294, Jun. 2005.
[23] W. Kempton, V. Udo, K. Huber, K. Komara, S. Letendre, S. Baker,
D. Brunner, and N. Pearre, A test of vehicle-to-grid (V2G) for energy
storage and frequency regulation in the PJM system, Univ. Delaware,
Newark, DE, USA, 2008.
[24] W. Kempton and J. Tomic, Vehicle-to-grid power fundamentals: Cal-
culating capacity and net revenue, J. Power Sources, vol. 144, no. 1,
pp. 268279, Jun. 2005.
[25] NYISO, Whole-sale Power. [Online]. Available: http://www.eia.gov/
cneaf/electricity/wholesale/new_yorkiso.html
[26] Demand Profile Generators. [Online]. Available: http://www.strath.ac.uk/
esru/
[27] Z. Ma, D. Callaway, and I. Hiskens, Decentralized charging control for
large populations of plug-in electric vehicles, in Proc. IEEE Int. Conf.
Control Appl., 2010, pp. 191195.
2927
[28] M. A. A. Pedrasa, T. D. Spooner, and I. F. MacGill, Coordinated schedul-
ing of residential distributed energy resources to optimize smart home
energy services, IEEE Trans. Smart Grid, vol. 1, no. 2, pp. 134143,
Sep. 2010.
[29] N. Ruiz, I. Cobelo, and J. Oyarzabal, A direct load control model for
virtual power plant management, IEEE Trans. Power Syst., vol. 24, no. 2,
pp. 959966, May 2009.
[30] A. Y. Saber and G. K. Venayagamoorthy, Plug-in vehicles and renew-
able energy sources for cost and emission reductions, IEEE Trans. Ind.
Electron., vol. 58, no. 4, pp. 12291238, Apr. 2011.
[31] S. Shao, M. Pipattanasomporn, and S. Rahman, Challenges of PHEV
penetration to the residential distribution network, in Proc. IEEE Power
Energy Soc. Gen. Meeting, 2009, pp. 18.
[32] S. Shao, T. Zhang, M. Pipattanasomporn, and S. Rahman, Impact of TOU
rates on distribution load shapes in a smart grid with PHEV penetration,
in Proc. IEEE PES Transmiss. Distrib. Conf. Expo., 2010, pp. 16.
[33] E. Sortomme, M. M. Hindi, S. D. J. MacPherson, and S. S. M.
Venkata, Sr., Coordinated charging of plug-in hybrid electric vehicles
to minimize distribution system losses, IEEE Trans. Smart Grid, vol. 2,
no. 1, pp. 198205, Mar. 2011.
[34] E. Sortomme and M. A. El-Sharkawi, Optimal charging strategies for
unidirectional vehicle-to-grid, IEEE Trans. Smart Grid, vol. 2, no. 1,
pp. 131138, Mar. 2011.
[35] C. D. White and K. M. Zhang, Using vehicle-to-grid technology for fre-
quency regulation and peak-load reduction, J. Power Sources, vol. 196,
no. 8, pp. 39723980, Apr. 2011.
[36] H. Zhang and M. Chow, Comprehensive dynamic battery modelling for
PHEV applications, in Proc. IEEE Power Energy Soc. Gen. Meeting,
2010, pp. 16.
Chenrui Jin (S10) received the B.S. degree in
physics and mathematics from Tsinghua University,
Beijing, China, in 2007 and the M.S.E.E. degree
from Syracuse University, Syracuse, NY, USA, in
2010. She is currently working toward the Ph.D. de-
gree with the Department of Electrical Engineering
and Computer Science, Syracuse University.
Her research interests include power system op-
timization, vehicle-to-grid technologies, distributed
energy sources, and energy management systems.
Jian Tang (M13) received the Ph.D. degree in com-
puter science from Arizona State University, Tempe,
AZ, USA, in 2006.
He is an Assistant Professor with the Department
of Electrical Engineering and Computer Science,
Syracuse University, Syracuse, NY, USA. His re-
search interests include wireless networking, cloud
computing, and green networking.
Dr. Tang received a National Science Foundation
CAREER Award in 2009. He served as a Cochair
for the IEEE Global Communications Conference
(GLOBECOM 2011) Wireless Networking Symposium, the International Con-
ference on Computing, Networking, and Communications (ICNC 20122013)
Wireless Networks Symposium, the Wireless and Optical Communications
Conference (WOCC 2013) Wireless Symposium, and the International Wire-
less Communications and Mobile Computing Conference (IWCMC 2013)
General Symposium. He has also served on the technical program committees
of many international conferences, such as the IEEE International Conference
on Computer Communications (Infocom), the IEEE International Conference
on Communications, and the IEEE GLOBECOM.
Prasanta Ghosh (SM11) is a Professor with the
Department of Electrical Engineering and Computer
Science, Syracuse University, Syracuse, NY, USA.
He conducts research in the areas of microelectronics
and power engineering. As a Fulbright Scholar, he
has traveled internationally to teach engineering stu-
dents and has delivered lectures on his research. His
current research includes smart grid, the design and
analysis of nanodevices, and thin-film gas sensors.